2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2008 The FreeBSD Foundation
5 * Copyright (c) 2009-2021 Bjoern A. Zeeb <bz@FreeBSD.org>
7 * This software was developed by CK Software GmbH under sponsorship
8 * from the FreeBSD Foundation.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * A pair of virtual back-to-back connected ethernet like interfaces
34 * (``two interfaces with a virtual cross-over cable'').
36 * This is mostly intended to be used to provide connectivity between
37 * different virtual network stack instances.
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
45 #include "opt_inet6.h"
47 #include <sys/param.h>
50 #include <sys/interrupt.h>
52 #include <sys/kernel.h>
53 #include <sys/libkern.h>
54 #include <sys/malloc.h>
56 #include <sys/module.h>
58 #include <sys/queue.h>
59 #include <sys/sched.h>
61 #include <sys/socket.h>
62 #include <sys/sockio.h>
63 #include <sys/taskqueue.h>
66 #include <net/ethernet.h>
68 #include <net/if_var.h>
69 #include <net/if_clone.h>
70 #include <net/if_media.h>
71 #include <net/if_var.h>
72 #include <net/if_types.h>
73 #include <net/netisr.h>
75 #include <net/rss_config.h>
77 #include <netinet/in_rss.h>
80 #include <netinet6/in6_rss.h>
85 static int epair_clone_match(struct if_clone *, const char *);
86 static int epair_clone_create(struct if_clone *, char *, size_t, caddr_t);
87 static int epair_clone_destroy(struct if_clone *, struct ifnet *);
89 static const char epairname[] = "epair";
90 #define RXRSIZE 4096 /* Probably overkill by 4-8x. */
92 static MALLOC_DEFINE(M_EPAIR, epairname,
93 "Pair of virtual cross-over connected Ethernet-like interfaces");
95 VNET_DEFINE_STATIC(struct if_clone *, epair_cloner);
96 #define V_epair_cloner VNET(epair_cloner)
98 static unsigned int next_index = 0;
99 #define EPAIR_LOCK_INIT() mtx_init(&epair_n_index_mtx, "epairidx", \
101 #define EPAIR_LOCK_DESTROY() mtx_destroy(&epair_n_index_mtx)
102 #define EPAIR_LOCK() mtx_lock(&epair_n_index_mtx)
103 #define EPAIR_UNLOCK() mtx_unlock(&epair_n_index_mtx)
116 struct epair_softc *sc;
119 static struct mtx epair_n_index_mtx;
121 struct ifnet *ifp; /* This ifp. */
122 struct ifnet *oifp; /* other ifp of pair. */
124 struct epair_queue *queues;
125 struct ifmedia media; /* Media config (fake). */
126 STAILQ_ENTRY(epair_softc) entry;
129 struct epair_tasks_t {
131 struct taskqueue *tq[MAXCPU];
134 static struct epair_tasks_t epair_tasks;
137 epair_clear_mbuf(struct mbuf *m)
139 /* Remove any CSUM_SND_TAG as ether_input will barf. */
140 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
141 m_snd_tag_rele(m->m_pkthdr.snd_tag);
142 m->m_pkthdr.snd_tag = NULL;
143 m->m_pkthdr.csum_flags &= ~CSUM_SND_TAG;
146 m_tag_delete_nonpersistent(m);
150 epair_tx_start_deferred(void *arg, int pending)
152 struct epair_queue *q = (struct epair_queue *)arg;
160 CURVNET_SET(ifp->if_vnet);
163 m = mbufq_flush(&q->q);
164 q->state = EPAIR_QUEUE_RUNNING;
168 n = STAILQ_NEXT(m, m_stailqpkt);
175 * Avoid flushing the queue more than once per task. We can otherwise
176 * end up starving ourselves in a multi-epair routing configuration.
179 if (mbufq_len(&q->q) > 0) {
181 q->state = EPAIR_QUEUE_WAKING;
184 q->state = EPAIR_QUEUE_IDLE;
189 taskqueue_enqueue(epair_tasks.tq[q->id], &q->tx_task);
195 static struct epair_queue *
196 epair_select_queue(struct epair_softc *sc, struct mbuf *m)
200 struct ether_header *eh;
203 ret = rss_m2bucket(m, &bucket);
205 /* Actually hash the packet. */
206 eh = mtod(m, struct ether_header *);
208 switch (ntohs(eh->ether_type)) {
211 rss_soft_m2cpuid_v4(m, 0, &bucket);
216 rss_soft_m2cpuid_v6(m, 0, &bucket);
224 bucket %= sc->num_queues;
228 return (&sc->queues[bucket]);
232 epair_prepare_mbuf(struct mbuf *m, struct ifnet *src_ifp)
236 if_setrcvif(m, src_ifp);
237 M_SETFIB(m, src_ifp->if_fib);
239 MPASS(m->m_nextpkt == NULL);
240 MPASS((m->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0);
244 epair_menq(struct mbuf *m, struct epair_softc *osc)
246 struct epair_queue *q;
247 struct ifnet *ifp, *oifp;
252 * I know this looks weird. We pass the "other sc" as we need that one
253 * and can get both ifps from it as well.
258 epair_prepare_mbuf(m, oifp);
260 /* Save values as once the mbuf is queued, it's not ours anymore. */
261 len = m->m_pkthdr.len;
262 mcast = (m->m_flags & (M_BCAST | M_MCAST)) != 0;
264 q = epair_select_queue(osc, m);
267 if (q->state == EPAIR_QUEUE_IDLE) {
268 q->state = EPAIR_QUEUE_WAKING;
269 taskqueue_enqueue(epair_tasks.tq[q->id], &q->tx_task);
271 error = mbufq_enqueue(&q->q, m);
276 if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1);
278 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
279 if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
281 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
282 if_inc_counter(oifp, IFCOUNTER_IPACKETS, 1);
287 epair_start(struct ifnet *ifp)
290 struct epair_softc *sc;
294 * We get packets here from ether_output via if_handoff()
295 * and need to put them into the input queue of the oifp
296 * and will put the packet into the receive-queue (rxq) of the
297 * other interface (oifp) of our pair.
303 IFQ_DEQUEUE(&ifp->if_snd, m);
309 /* In case either interface is not usable drop the packet. */
310 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
311 (ifp->if_flags & IFF_UP) == 0 ||
312 (oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
313 (oifp->if_flags & IFF_UP) == 0) {
318 (void) epair_menq(m, sc);
323 epair_transmit(struct ifnet *ifp, struct mbuf *m)
325 struct epair_softc *sc;
337 * We are not going to use the interface en/dequeue mechanism
338 * on the TX side. We are called from ether_output_frame()
339 * and will put the packet into the receive-queue (rxq) of the
340 * other interface (oifp) of our pair.
342 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
344 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
347 if ((ifp->if_flags & IFF_UP) == 0) {
349 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
356 * In case the outgoing interface is not usable,
361 if ((oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
362 (oifp->if_flags & IFF_UP) == 0) {
363 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
369 len = m->m_pkthdr.len;
370 mcast = (m->m_flags & (M_BCAST | M_MCAST)) != 0;
373 /* Support ALTQ via the classic if_start() path. */
374 IF_LOCK(&ifp->if_snd);
375 if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
376 ALTQ_ENQUEUE(&ifp->if_snd, m, NULL, error);
378 if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1);
379 IF_UNLOCK(&ifp->if_snd);
381 if_inc_counter(ifp, IFCOUNTER_OBYTES, len);
383 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
388 IF_UNLOCK(&ifp->if_snd);
391 epair_menq(m, oifp->if_softc);
396 epair_media_change(struct ifnet *ifp __unused)
404 epair_media_status(struct ifnet *ifp __unused, struct ifmediareq *imr)
407 imr->ifm_status = IFM_AVALID | IFM_ACTIVE;
408 imr->ifm_active = IFM_ETHER | IFM_10G_T | IFM_FDX;
412 epair_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
414 struct epair_softc *sc;
418 ifr = (struct ifreq *)data;
429 error = ifmedia_ioctl(ifp, ifr, &sc->media, cmd);
433 /* We basically allow all kinds of MTUs. */
434 ifp->if_mtu = ifr->ifr_mtu;
439 /* Let the common ethernet handler process this. */
440 error = ether_ioctl(ifp, cmd, data);
448 epair_init(void *dummy __unused)
453 * Interface cloning functions.
454 * We use our private ones so that we can create/destroy our secondary
455 * device along with the primary one.
458 epair_clone_match(struct if_clone *ifc, const char *name)
463 * Our base name is epair.
464 * Our interfaces will be named epair<n>[ab].
465 * So accept anything of the following list:
468 * but not the epair<n>[ab] versions.
470 if (strncmp(epairname, name, sizeof(epairname)-1) != 0)
473 for (cp = name + sizeof(epairname) - 1; *cp != '\0'; cp++) {
474 if (*cp < '0' || *cp > '9')
482 epair_clone_add(struct if_clone *ifc, struct epair_softc *scb)
485 uint8_t eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
488 /* Copy epairNa etheraddr and change the last byte. */
489 memcpy(eaddr, scb->oifp->if_hw_addr, ETHER_ADDR_LEN);
491 ether_ifattach(ifp, eaddr);
493 if_clone_addif(ifc, ifp);
497 epair_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
499 struct epair_softc *sca, *scb;
502 int error, unit, wildcard;
506 uint8_t eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
508 /* Try to see if a special unit was requested. */
509 error = ifc_name2unit(name, &unit);
512 wildcard = (unit < 0);
514 error = ifc_alloc_unit(ifc, &unit);
519 * If no unit had been given, we need to adjust the ifName.
520 * Also make sure there is space for our extra [ab] suffix.
522 for (dp = name; *dp != '\0'; dp++);
524 error = snprintf(dp, len - (dp - name), "%d", unit);
525 if (error > len - (dp - name) - 1) {
526 /* ifName too long. */
527 ifc_free_unit(ifc, unit);
532 if (len - (dp - name) - 1 < 1) {
533 /* No space left for our [ab] suffix. */
534 ifc_free_unit(ifc, unit);
538 /* Must not change dp so we can replace 'a' by 'b' later. */
541 /* Check if 'a' and 'b' interfaces already exist. */
542 if (ifunit(name) != NULL)
545 if (ifunit(name) != NULL)
548 /* Allocate memory for both [ab] interfaces */
549 sca = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO);
550 sca->ifp = if_alloc(IFT_ETHER);
551 sca->num_queues = epair_tasks.tasks;
552 if (sca->ifp == NULL) {
554 ifc_free_unit(ifc, unit);
557 sca->queues = mallocarray(sca->num_queues, sizeof(struct epair_queue),
559 for (int i = 0; i < sca->num_queues; i++) {
560 struct epair_queue *q = &sca->queues[i];
562 q->state = EPAIR_QUEUE_IDLE;
563 mtx_init(&q->mtx, "epaiq", NULL, MTX_DEF | MTX_NEW);
564 mbufq_init(&q->q, RXRSIZE);
566 NET_TASK_INIT(&q->tx_task, 0, epair_tx_start_deferred, q);
569 scb = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO);
570 scb->ifp = if_alloc(IFT_ETHER);
571 scb->num_queues = epair_tasks.tasks;
572 if (scb->ifp == NULL) {
576 ifc_free_unit(ifc, unit);
579 scb->queues = mallocarray(scb->num_queues, sizeof(struct epair_queue),
581 for (int i = 0; i < scb->num_queues; i++) {
582 struct epair_queue *q = &scb->queues[i];
584 q->state = EPAIR_QUEUE_IDLE;
585 mtx_init(&q->mtx, "epaiq", NULL, MTX_DEF | MTX_NEW);
586 mbufq_init(&q->q, RXRSIZE);
588 NET_TASK_INIT(&q->tx_task, 0, epair_tx_start_deferred, q);
592 * Cross-reference the interfaces so we will be able to free both.
594 sca->oifp = scb->ifp;
595 scb->oifp = sca->ifp;
599 /* Get an approximate distribution. */
600 hash = next_index % mp_ncpus;
606 /* Initialise pseudo media types. */
607 ifmedia_init(&sca->media, 0, epair_media_change, epair_media_status);
608 ifmedia_add(&sca->media, IFM_ETHER | IFM_10G_T, 0, NULL);
609 ifmedia_set(&sca->media, IFM_ETHER | IFM_10G_T);
610 ifmedia_init(&scb->media, 0, epair_media_change, epair_media_status);
611 ifmedia_add(&scb->media, IFM_ETHER | IFM_10G_T, 0, NULL);
612 ifmedia_set(&scb->media, IFM_ETHER | IFM_10G_T);
614 /* Finish initialization of interface <n>a. */
617 strlcpy(ifp->if_xname, name, IFNAMSIZ);
618 ifp->if_dname = epairname;
619 ifp->if_dunit = unit;
620 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
621 ifp->if_flags |= IFF_KNOWSEPOCH;
622 ifp->if_capabilities = IFCAP_VLAN_MTU;
623 ifp->if_capenable = IFCAP_VLAN_MTU;
624 ifp->if_start = epair_start;
625 ifp->if_ioctl = epair_ioctl;
626 ifp->if_init = epair_init;
627 if_setsendqlen(ifp, ifqmaxlen);
628 if_setsendqready(ifp);
631 * Calculate the etheraddr hashing the hostid and the
632 * interface index. The result would be hopefully unique.
633 * Note that the "a" component of an epair instance may get moved
634 * to a different VNET after creation. In that case its index
635 * will be freed and the index can get reused by new epair instance.
636 * Make sure we do not create same etheraddr again.
638 getcredhostid(curthread->td_ucred, (unsigned long *)&hostid);
640 arc4rand(&hostid, sizeof(hostid), 0);
643 if (ifp->if_index > next_index)
644 next_index = ifp->if_index;
648 key[0] = (uint32_t)next_index;
650 key[1] = (uint32_t)(hostid & 0xffffffff);
651 key[2] = (uint32_t)((hostid >> 32) & 0xfffffffff);
652 hash = jenkins_hash32(key, 3, 0);
655 memcpy(&eaddr[1], &hash, 4);
657 ether_ifattach(ifp, eaddr);
658 ifp->if_baudrate = IF_Gbps(10); /* arbitrary maximum */
659 ifp->if_transmit = epair_transmit;
661 /* Swap the name and finish initialization of interface <n>b. */
666 strlcpy(ifp->if_xname, name, IFNAMSIZ);
667 ifp->if_dname = epairname;
668 ifp->if_dunit = unit;
669 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
670 ifp->if_flags |= IFF_KNOWSEPOCH;
671 ifp->if_capabilities = IFCAP_VLAN_MTU;
672 ifp->if_capenable = IFCAP_VLAN_MTU;
673 ifp->if_start = epair_start;
674 ifp->if_ioctl = epair_ioctl;
675 ifp->if_init = epair_init;
676 if_setsendqlen(ifp, ifqmaxlen);
677 if_setsendqready(ifp);
678 /* We need to play some tricks here for the second interface. */
679 strlcpy(name, epairname, len);
681 /* Correctly set the name for the cloner list. */
682 strlcpy(name, scb->ifp->if_xname, len);
683 epair_clone_add(ifc, scb);
685 ifp->if_baudrate = IF_Gbps(10); /* arbitrary maximum */
686 ifp->if_transmit = epair_transmit;
689 * Restore name to <n>a as the ifp for this will go into the
690 * cloner list for the initial call.
692 strlcpy(name, sca->ifp->if_xname, len);
694 /* Tell the world, that we are ready to rock. */
695 sca->ifp->if_drv_flags |= IFF_DRV_RUNNING;
696 if_link_state_change(sca->ifp, LINK_STATE_UP);
697 scb->ifp->if_drv_flags |= IFF_DRV_RUNNING;
698 if_link_state_change(scb->ifp, LINK_STATE_UP);
704 epair_drain_rings(struct epair_softc *sc)
706 for (int i = 0; i < sc->num_queues; i++) {
707 struct epair_queue *q;
712 m = mbufq_flush(&q->q);
715 for (; m != NULL; m = n) {
723 epair_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
726 struct epair_softc *sca, *scb;
730 * In case we called into if_clone_destroyif() ourselves
731 * again to remove the second interface, the softc will be
732 * NULL. In that case so not do anything but return success.
734 if (ifp->if_softc == NULL)
737 unit = ifp->if_dunit;
740 scb = oifp->if_softc;
742 /* Frist get the interfaces down and detached. */
743 if_link_state_change(ifp, LINK_STATE_DOWN);
744 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
745 if_link_state_change(oifp, LINK_STATE_DOWN);
746 oifp->if_drv_flags &= ~IFF_DRV_RUNNING;
749 ether_ifdetach(oifp);
751 /* Third free any queued packets and all the resources. */
752 CURVNET_SET_QUIET(oifp->if_vnet);
753 epair_drain_rings(scb);
754 oifp->if_softc = NULL;
755 error = if_clone_destroyif(ifc, oifp);
757 panic("%s: if_clone_destroyif() for our 2nd iface failed: %d",
760 ifmedia_removeall(&scb->media);
761 for (int i = 0; i < scb->num_queues; i++) {
762 struct epair_queue *q = &scb->queues[i];
763 mtx_destroy(&q->mtx);
765 free(scb->queues, M_EPAIR);
769 epair_drain_rings(sca);
771 ifmedia_removeall(&sca->media);
772 for (int i = 0; i < sca->num_queues; i++) {
773 struct epair_queue *q = &sca->queues[i];
774 mtx_destroy(&q->mtx);
776 free(sca->queues, M_EPAIR);
779 /* Last free the cloner unit. */
780 ifc_free_unit(ifc, unit);
786 vnet_epair_init(const void *unused __unused)
789 V_epair_cloner = if_clone_advanced(epairname, 0,
790 epair_clone_match, epair_clone_create, epair_clone_destroy);
792 VNET_SYSINIT(vnet_epair_init, SI_SUB_PSEUDO, SI_ORDER_ANY,
793 vnet_epair_init, NULL);
796 vnet_epair_uninit(const void *unused __unused)
799 if_clone_detach(V_epair_cloner);
801 VNET_SYSUNINIT(vnet_epair_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY,
802 vnet_epair_uninit, NULL);
808 epair_tasks.tasks = 0;
816 /* Pin to this CPU so we get appropriate NUMA allocations. */
817 thread_lock(curthread);
818 sched_bind(curthread, cpu);
819 thread_unlock(curthread);
821 snprintf(name, sizeof(name), "epair_task_%d", cpu);
823 epair_tasks.tq[cpu] = taskqueue_create(name, M_WAITOK,
824 taskqueue_thread_enqueue,
825 &epair_tasks.tq[cpu]);
826 CPU_SETOF(cpu, &cpu_mask);
827 taskqueue_start_threads_cpuset(&epair_tasks.tq[cpu], 1, PI_NET,
828 &cpu_mask, "%s", name);
832 thread_lock(curthread);
833 sched_unbind(curthread);
834 thread_unlock(curthread);
836 snprintf(name, sizeof(name), "epair_task");
838 epair_tasks.tq[0] = taskqueue_create(name, M_WAITOK,
839 taskqueue_thread_enqueue,
841 taskqueue_start_threads(&epair_tasks.tq[0], 1, PI_NET, "%s", name);
843 epair_tasks.tasks = 1;
853 for (int i = 0; i < epair_tasks.tasks; i++) {
854 taskqueue_drain_all(epair_tasks.tq[i]);
855 taskqueue_free(epair_tasks.tq[i]);
860 epair_modevent(module_t mod, int type, void *data)
867 ret = epair_mod_init();
871 printf("%s: %s initialized.\n", __func__, epairname);
875 EPAIR_LOCK_DESTROY();
877 printf("%s: %s unloaded.\n", __func__, epairname);
885 static moduledata_t epair_mod = {
891 DECLARE_MODULE(if_epair, epair_mod, SI_SUB_PSEUDO, SI_ORDER_MIDDLE);
892 MODULE_VERSION(if_epair, 3);